monai-weekly 1.5.dev2444__py3-none-any.whl → 1.5.dev2446__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
monai/__init__.py CHANGED
@@ -136,4 +136,4 @@ except BaseException:
136
136
 
137
137
  if MONAIEnvVars.debug():
138
138
  raise
139
- __commit_id__ = "c1ceea3d4cbb0781eae4e209b80fe651a776fed2"
139
+ __commit_id__ = "13b96aedc48ad2da16149490b06a1a6bd8361335"
monai/_version.py CHANGED
@@ -8,11 +8,11 @@ import json
8
8
 
9
9
  version_json = '''
10
10
  {
11
- "date": "2024-11-03T02:29:11+0000",
11
+ "date": "2024-11-17T02:30:32+0000",
12
12
  "dirty": false,
13
13
  "error": null,
14
- "full-revisionid": "c70fbd8ff919cabaacfabcbdbf28aa435ae622f9",
15
- "version": "1.5.dev2444"
14
+ "full-revisionid": "218216250ce297265400abe56ee915898d75a2ec",
15
+ "version": "1.5.dev2446"
16
16
  }
17
17
  ''' # END VERSION_JSON
18
18
 
monai/bundle/scripts.py CHANGED
@@ -1589,6 +1589,8 @@ def trt_export(
1589
1589
  """
1590
1590
  Export the model checkpoint to the given filepath as a TensorRT engine-based TorchScript.
1591
1591
  Currently, this API only supports converting models whose inputs are all tensors.
1592
+ Note: NVIDIA Volta support (GPUs with compute capability 7.0) has been removed starting with TensorRT 10.5.
1593
+ Review the TensorRT Support Matrix for which GPUs are supported.
1592
1594
 
1593
1595
  There are two ways to export a model:
1594
1596
  1, Torch-TensorRT way: PyTorch module ---> TorchScript module ---> TensorRT engine-based TorchScript.
@@ -26,6 +26,7 @@ from .encoder import BaseEncoder
26
26
  from .fcn import FCN, GCN, MCFCN, Refine
27
27
  from .feature_pyramid_network import ExtraFPNBlock, FeaturePyramidNetwork, LastLevelMaxPool, LastLevelP6P7
28
28
  from .localnet_block import LocalNetDownSampleBlock, LocalNetFeatureExtractorBlock, LocalNetUpSampleBlock
29
+ from .mednext_block import MedNeXtBlock, MedNeXtDownBlock, MedNeXtOutBlock, MedNeXtUpBlock
29
30
  from .mlp import MLPBlock
30
31
  from .patchembedding import PatchEmbed, PatchEmbeddingBlock
31
32
  from .regunet_block import RegistrationDownSampleBlock, RegistrationExtractionBlock, RegistrationResidualConvBlock
@@ -0,0 +1,309 @@
1
+ # Copyright (c) MONAI Consortium
2
+ # Licensed under the Apache License, Version 2.0 (the "License");
3
+ # you may not use this file except in compliance with the License.
4
+ # You may obtain a copy of the License at
5
+ # http://www.apache.org/licenses/LICENSE-2.0
6
+ # Unless required by applicable law or agreed to in writing, software
7
+ # distributed under the License is distributed on an "AS IS" BASIS,
8
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
9
+ # See the License for the specific language governing permissions and
10
+ # limitations under the License.
11
+
12
+ # Portions of this code are derived from the original repository at:
13
+ # https://github.com/MIC-DKFZ/MedNeXt
14
+ # and are used under the terms of the Apache License, Version 2.0.
15
+
16
+ from __future__ import annotations
17
+
18
+ import torch
19
+ import torch.nn as nn
20
+
21
+ all = ["MedNeXtBlock", "MedNeXtDownBlock", "MedNeXtUpBlock", "MedNeXtOutBlock"]
22
+
23
+
24
+ def get_conv_layer(spatial_dim: int = 3, transpose: bool = False):
25
+ if spatial_dim == 2:
26
+ return nn.ConvTranspose2d if transpose else nn.Conv2d
27
+ else: # spatial_dim == 3
28
+ return nn.ConvTranspose3d if transpose else nn.Conv3d
29
+
30
+
31
+ class MedNeXtBlock(nn.Module):
32
+ """
33
+ MedNeXtBlock class for the MedNeXt model.
34
+
35
+ Args:
36
+ in_channels (int): Number of input channels.
37
+ out_channels (int): Number of output channels.
38
+ expansion_ratio (int): Expansion ratio for the block. Defaults to 4.
39
+ kernel_size (int): Kernel size for convolutions. Defaults to 7.
40
+ use_residual_connection (int): Whether to use residual connection. Defaults to True.
41
+ norm_type (str): Type of normalization to use. Defaults to "group".
42
+ dim (str): Dimension of the input. Can be "2d" or "3d". Defaults to "3d".
43
+ global_resp_norm (bool): Whether to use global response normalization. Defaults to False.
44
+ """
45
+
46
+ def __init__(
47
+ self,
48
+ in_channels: int,
49
+ out_channels: int,
50
+ expansion_ratio: int = 4,
51
+ kernel_size: int = 7,
52
+ use_residual_connection: int = True,
53
+ norm_type: str = "group",
54
+ dim="3d",
55
+ global_resp_norm=False,
56
+ ):
57
+
58
+ super().__init__()
59
+
60
+ self.do_res = use_residual_connection
61
+
62
+ self.dim = dim
63
+ conv = get_conv_layer(spatial_dim=2 if dim == "2d" else 3)
64
+ global_resp_norm_param_shape = (1,) * (2 if dim == "2d" else 3)
65
+ # First convolution layer with DepthWise Convolutions
66
+ self.conv1 = conv(
67
+ in_channels=in_channels,
68
+ out_channels=in_channels,
69
+ kernel_size=kernel_size,
70
+ stride=1,
71
+ padding=kernel_size // 2,
72
+ groups=in_channels,
73
+ )
74
+
75
+ # Normalization Layer. GroupNorm is used by default.
76
+ if norm_type == "group":
77
+ self.norm = nn.GroupNorm(num_groups=in_channels, num_channels=in_channels) # type: ignore
78
+ elif norm_type == "layer":
79
+ self.norm = nn.LayerNorm(
80
+ normalized_shape=[in_channels] + [kernel_size] * (2 if dim == "2d" else 3) # type: ignore
81
+ )
82
+ # Second convolution (Expansion) layer with Conv3D 1x1x1
83
+ self.conv2 = conv(
84
+ in_channels=in_channels, out_channels=expansion_ratio * in_channels, kernel_size=1, stride=1, padding=0
85
+ )
86
+
87
+ # GeLU activations
88
+ self.act = nn.GELU()
89
+
90
+ # Third convolution (Compression) layer with Conv3D 1x1x1
91
+ self.conv3 = conv(
92
+ in_channels=expansion_ratio * in_channels, out_channels=out_channels, kernel_size=1, stride=1, padding=0
93
+ )
94
+
95
+ self.global_resp_norm = global_resp_norm
96
+ if self.global_resp_norm:
97
+ global_resp_norm_param_shape = (1, expansion_ratio * in_channels) + global_resp_norm_param_shape
98
+ self.global_resp_beta = nn.Parameter(torch.zeros(global_resp_norm_param_shape), requires_grad=True)
99
+ self.global_resp_gamma = nn.Parameter(torch.zeros(global_resp_norm_param_shape), requires_grad=True)
100
+
101
+ def forward(self, x):
102
+ """
103
+ Forward pass of the MedNeXtBlock.
104
+
105
+ Args:
106
+ x (torch.Tensor): Input tensor.
107
+
108
+ Returns:
109
+ torch.Tensor: Output tensor.
110
+ """
111
+ x1 = x
112
+ x1 = self.conv1(x1)
113
+ x1 = self.act(self.conv2(self.norm(x1)))
114
+
115
+ if self.global_resp_norm:
116
+ # gamma, beta: learnable affine transform parameters
117
+ # X: input of shape (N,C,H,W,D)
118
+ if self.dim == "2d":
119
+ gx = torch.norm(x1, p=2, dim=(-2, -1), keepdim=True)
120
+ else:
121
+ gx = torch.norm(x1, p=2, dim=(-3, -2, -1), keepdim=True)
122
+ nx = gx / (gx.mean(dim=1, keepdim=True) + 1e-6)
123
+ x1 = self.global_resp_gamma * (x1 * nx) + self.global_resp_beta + x1
124
+ x1 = self.conv3(x1)
125
+ if self.do_res:
126
+ x1 = x + x1
127
+ return x1
128
+
129
+
130
+ class MedNeXtDownBlock(MedNeXtBlock):
131
+ """
132
+ MedNeXtDownBlock class for downsampling in the MedNeXt model.
133
+
134
+ Args:
135
+ in_channels (int): Number of input channels.
136
+ out_channels (int): Number of output channels.
137
+ expansion_ratio (int): Expansion ratio for the block. Defaults to 4.
138
+ kernel_size (int): Kernel size for convolutions. Defaults to 7.
139
+ use_residual_connection (bool): Whether to use residual connection. Defaults to False.
140
+ norm_type (str): Type of normalization to use. Defaults to "group".
141
+ dim (str): Dimension of the input. Can be "2d" or "3d". Defaults to "3d".
142
+ global_resp_norm (bool): Whether to use global response normalization. Defaults to False.
143
+ """
144
+
145
+ def __init__(
146
+ self,
147
+ in_channels: int,
148
+ out_channels: int,
149
+ expansion_ratio: int = 4,
150
+ kernel_size: int = 7,
151
+ use_residual_connection: bool = False,
152
+ norm_type: str = "group",
153
+ dim: str = "3d",
154
+ global_resp_norm: bool = False,
155
+ ):
156
+
157
+ super().__init__(
158
+ in_channels,
159
+ out_channels,
160
+ expansion_ratio,
161
+ kernel_size,
162
+ use_residual_connection=False,
163
+ norm_type=norm_type,
164
+ dim=dim,
165
+ global_resp_norm=global_resp_norm,
166
+ )
167
+
168
+ conv = get_conv_layer(spatial_dim=2 if dim == "2d" else 3)
169
+ self.resample_do_res = use_residual_connection
170
+ if use_residual_connection:
171
+ self.res_conv = conv(in_channels=in_channels, out_channels=out_channels, kernel_size=1, stride=2)
172
+
173
+ self.conv1 = conv(
174
+ in_channels=in_channels,
175
+ out_channels=in_channels,
176
+ kernel_size=kernel_size,
177
+ stride=2,
178
+ padding=kernel_size // 2,
179
+ groups=in_channels,
180
+ )
181
+
182
+ def forward(self, x):
183
+ """
184
+ Forward pass of the MedNeXtDownBlock.
185
+
186
+ Args:
187
+ x (torch.Tensor): Input tensor.
188
+
189
+ Returns:
190
+ torch.Tensor: Output tensor.
191
+ """
192
+ x1 = super().forward(x)
193
+
194
+ if self.resample_do_res:
195
+ res = self.res_conv(x)
196
+ x1 = x1 + res
197
+
198
+ return x1
199
+
200
+
201
+ class MedNeXtUpBlock(MedNeXtBlock):
202
+ """
203
+ MedNeXtUpBlock class for upsampling in the MedNeXt model.
204
+
205
+ Args:
206
+ in_channels (int): Number of input channels.
207
+ out_channels (int): Number of output channels.
208
+ expansion_ratio (int): Expansion ratio for the block. Defaults to 4.
209
+ kernel_size (int): Kernel size for convolutions. Defaults to 7.
210
+ use_residual_connection (bool): Whether to use residual connection. Defaults to False.
211
+ norm_type (str): Type of normalization to use. Defaults to "group".
212
+ dim (str): Dimension of the input. Can be "2d" or "3d". Defaults to "3d".
213
+ global_resp_norm (bool): Whether to use global response normalization. Defaults to False.
214
+ """
215
+
216
+ def __init__(
217
+ self,
218
+ in_channels: int,
219
+ out_channels: int,
220
+ expansion_ratio: int = 4,
221
+ kernel_size: int = 7,
222
+ use_residual_connection: bool = False,
223
+ norm_type: str = "group",
224
+ dim: str = "3d",
225
+ global_resp_norm: bool = False,
226
+ ):
227
+ super().__init__(
228
+ in_channels,
229
+ out_channels,
230
+ expansion_ratio,
231
+ kernel_size,
232
+ use_residual_connection=False,
233
+ norm_type=norm_type,
234
+ dim=dim,
235
+ global_resp_norm=global_resp_norm,
236
+ )
237
+
238
+ self.resample_do_res = use_residual_connection
239
+
240
+ self.dim = dim
241
+ conv = get_conv_layer(spatial_dim=2 if dim == "2d" else 3, transpose=True)
242
+ if use_residual_connection:
243
+ self.res_conv = conv(in_channels=in_channels, out_channels=out_channels, kernel_size=1, stride=2)
244
+
245
+ self.conv1 = conv(
246
+ in_channels=in_channels,
247
+ out_channels=in_channels,
248
+ kernel_size=kernel_size,
249
+ stride=2,
250
+ padding=kernel_size // 2,
251
+ groups=in_channels,
252
+ )
253
+
254
+ def forward(self, x):
255
+ """
256
+ Forward pass of the MedNeXtUpBlock.
257
+
258
+ Args:
259
+ x (torch.Tensor): Input tensor.
260
+
261
+ Returns:
262
+ torch.Tensor: Output tensor.
263
+ """
264
+ x1 = super().forward(x)
265
+ # Asymmetry but necessary to match shape
266
+
267
+ if self.dim == "2d":
268
+ x1 = torch.nn.functional.pad(x1, (1, 0, 1, 0))
269
+ else:
270
+ x1 = torch.nn.functional.pad(x1, (1, 0, 1, 0, 1, 0))
271
+
272
+ if self.resample_do_res:
273
+ res = self.res_conv(x)
274
+ if self.dim == "2d":
275
+ res = torch.nn.functional.pad(res, (1, 0, 1, 0))
276
+ else:
277
+ res = torch.nn.functional.pad(res, (1, 0, 1, 0, 1, 0))
278
+ x1 = x1 + res
279
+
280
+ return x1
281
+
282
+
283
+ class MedNeXtOutBlock(nn.Module):
284
+ """
285
+ MedNeXtOutBlock class for the output block in the MedNeXt model.
286
+
287
+ Args:
288
+ in_channels (int): Number of input channels.
289
+ n_classes (int): Number of output classes.
290
+ dim (str): Dimension of the input. Can be "2d" or "3d".
291
+ """
292
+
293
+ def __init__(self, in_channels, n_classes, dim):
294
+ super().__init__()
295
+
296
+ conv = get_conv_layer(spatial_dim=2 if dim == "2d" else 3, transpose=True)
297
+ self.conv_out = conv(in_channels, n_classes, kernel_size=1)
298
+
299
+ def forward(self, x):
300
+ """
301
+ Forward pass of the MedNeXtOutBlock.
302
+
303
+ Args:
304
+ x (torch.Tensor): Input tensor.
305
+
306
+ Returns:
307
+ torch.Tensor: Output tensor.
308
+ """
309
+ return self.conv_out(x)
@@ -53,6 +53,25 @@ from .fullyconnectednet import FullyConnectedNet, VarFullyConnectedNet
53
53
  from .generator import Generator
54
54
  from .highresnet import HighResBlock, HighResNet
55
55
  from .hovernet import Hovernet, HoVernet, HoVerNet, HoverNet
56
+ from .mednext import (
57
+ MedNeXt,
58
+ MedNext,
59
+ MedNextB,
60
+ MedNeXtB,
61
+ MedNextBase,
62
+ MedNextL,
63
+ MedNeXtL,
64
+ MedNeXtLarge,
65
+ MedNextLarge,
66
+ MedNextM,
67
+ MedNeXtM,
68
+ MedNeXtMedium,
69
+ MedNextMedium,
70
+ MedNextS,
71
+ MedNeXtS,
72
+ MedNeXtSmall,
73
+ MedNextSmall,
74
+ )
56
75
  from .milmodel import MILModel
57
76
  from .netadapter import NetAdapter
58
77
  from .patchgan_discriminator import MultiScalePatchDiscriminator, PatchDiscriminator